Method · Maple Labs · Updated October 2026
The Oracle Method: conversational analysis
Oracle, the tool built by Maple Labs, our technology partner, does not track a list of prompts. It replays the buying situations of a market: agents play prospects and talk to ChatGPT, Claude and Gemini, while an Analyst agent records what the AI searches, reads, compares and recommends. Here is the method, what it delivers, and what it does not claim.
What the Oracle Method measures
The Oracle Method, developed by Maple Labs, measures whether AI assistants recommend a brand, and why. It does not replay a list of prompts: it reproduces the buying situations of a market and observes how the AI clarifies the need, discovers options, checks them and recommends one.
This page covers, in order: the conversations, what Oracle extracts from them (decision factors, discovery and coverage sources, real competitors), the diagnosis, the assistants, how to read the results, and the limits Maple Labs publishes.
Botanik documents it because a number without a method is an opinion. With the method in the open, you can check what a status means before you buy an audit.
The key facts, at a glance. Scale: more than 2,500 AI searches analysed in a single audit. Assistants: ChatGPT, Claude and Gemini. Readout: absent, mentioned or recommended, factor by factor.
Buying situations, not a list of prompts
A measurement that starts from a list of prompts measures the questions someone imagined. Oracle starts from a buying situation: an industry, a commercial scope (country, city, online or physical network) and a category of offers that meet the same need. Not the whole company, not a keyword list.
Within that frame, Replica agents play prospects, each with its own needs, hesitations and constraints, moving the search forward like a real buyer. An Analyst agent observes each conversation without taking part, and records what the assistant searches for, the pages it opens, the options it compares and the one it recommends.
Players and sources emerge from the exchanges: nobody preselects them. In a recent audit, more than 2,500 AI searches were analysed.
Nothing here claims to be your customers' real prompts: Maple Labs says plainly that nobody has them. The conversations are simulated, and built to behave like buyers.
What Oracle extracts: factors, sources, competitors
The dimensions on which the AI compares offers, in three families: expectations (reviews, user feedback), trends (market, regulation) and criteria (requirements, standards). One recent audit surfaced 30.
The pages that bring up the companies the AI considers: comparisons, rankings, directories, communities. If you're not there, you're not on its list.
The pages where the AI looks for what qualifies each player, factor by factor: 8 to 20 per factor, of which the five most recurrent are kept. Without that proof, it may cite you, but it won't recommend you.
We give Oracle no competitors. The AI finds the ones it compares, including specialists, local players or indirect competitors you weren't watching.
From the conversations, Oracle extracts four things, consolidated into one reference grid for your market.
The diagnosis: explore the market, then go deeper factor by factor
The first conversations surface the players, sources and decision factors of the market. Targeted conversations then go deeper on each factor by making it explicit in the request: they show which players come up on that criterion, which are only mentioned, and which one the AI recommends.
For every page, Oracle tracks how far the information got: found, opened, extracted, tied to a company and a criterion, then used in the comparison.
That is what locates the problem when your brand is missing: the offer (the expected capability doesn't exist), understanding (it exists but is poorly explained or scattered), or information visibility (the proof exists, but the AI can't find it or can't read it).
That is why the audit doesn't end on a percentage: it ends on a list of factors and sources to work on.
Assistants: ChatGPT, Claude and Gemini, on a common basis
Oracle works on ChatGPT, Claude and Gemini. According to Maple Labs, these three assistants cover 99% of consumer scenarios in the scope studied. Their searches, sources and answers vary, but they follow the same logic to recommend: clarify the need, discover options, check them, compare them.
Differences between assistants that would change the conclusions are built into the analysis. The rest does not justify an engine-by-engine verdict, which is why Botanik does not sell one.
AI answers move with market news and model updates, so every audit is dated, and re-run, usually every quarter.
Absent, mentioned, recommended: factor by factor
A brand is mentioned when it appears in the answer. It is recommended when the AI explicitly presents it as the option that fits the request, or the preferable one. That preference is what counts, not a place in a list, and it reads factor by factor, against the competitors the AI actually considers.
| State | What it means | What it changes | Horizon de résultat |
|---|---|---|---|
| Absent | The AI does not cite you on this factor | You don't exist in the buying conversation. Your competitor does. | |
| Mentioned | Cited, with no preference | You're on the list, but the AI doesn't push you forward. | |
| Recommended | Presented as the right choice | That's where the traffic and the leads go. |
What Oracle does not claim
Nobody has the prompts people actually type into AI assistants. Oracle replays buying situations, and says so.
The audit shows what is missing. It does not promise you will be recommended in every context.
Oracle does not try to game AI answers. The work is on what the AI reads: your pages and the sources it trusts.
A page consulted during a search does not tell whether it was used to train the model. Oracle adds an AI layer on top of SEO; it does not replace it, and it does not predict every answer.
An honest method writes down its limits. Maple Labs publishes its own, and Botanik repeats them as they are.
One method, one audit
At Botanik, the Oracle Method runs inside the GEO audit, from €2,670 one-shot, for one product range, three weeks from framing to debrief.
The audit ends with an action plan: for each missing factor, the exact page to enrich, on your site or a third party's, and the text to add, already written.
After the audit, AI traffic and AI accessibility are tracked every month in your reporting, and the audit is re-run, usually every quarter.
How to cite the Oracle Method
The Oracle Method is developed by Maple Labs. This page describes what it does and how Botanik applies it; it is the English version of botanik.ai/methode-oracle.
If you quote a figure from an audit, keep its date attached: AI answers move, and a status only holds for the period it was measured.
If you quote this page, credit Maple Labs for the Oracle Method and link to this page for how Botanik applies it.
Does the Oracle Method use a list of prompts?
No. It replays buying situations: Replica agents play prospects and talk to the assistants, and the topics, players and sources emerge from those conversations. Targeted conversations then go deeper on each decision factor. In a single audit, more than 2,500 AI searches are analysed.
Which assistants does the Oracle Method study?
ChatGPT, Claude and Gemini, which according to Maple Labs cover 99% of consumer scenarios in the scope studied. Differences that would change the conclusions are built into the analysis, so the audit does not deliver an engine-by-engine verdict.
How does Oracle deal with unstable AI answers?
It works on large volumes of comparable conversations and keeps what recurs from one to the next, rather than a single answer. Audits are dated and re-run, usually every quarter, because answers move with the market and with model updates.
What does Recommended mean in the Oracle Method?
Recommended means the AI explicitly presents your brand as the option that fits the request, or the preferable one, on a given decision factor. Mentioned means it cites you without that preference. Absent means it does not cite you on that factor.
Can I reuse the Oracle Method?
The Oracle Method belongs to Maple Labs, so it is not Botanik's to license. You can quote this page with a link and credit Maple Labs for the method. This page is the English version of botanik.ai/methode-oracle.
How much does the Oracle audit cost?
The Oracle audit starts at €2,670 one-shot, for one product range, delivered in three weeks, with the action plan already written. The full price grid is public.
Case studies, measured results
How Pennylane strengthened its SEO presence on key accounting and financial queries
With Botanik, Pennylane structured a comprehensive SEO strategy combining technical optimization, educational content, and semantic targeting to strengthen its position in the financial management tools ecosystem.
Franck Neuenschwander · founder of Botanik
The method you just read, Organic Inference™, comes from Maple Labs. What's mine is how we apply it, and I answer for it. 15 years in-house, on the brand side, and the only agency I've ever worked at is the one I run: if you want to discuss it, it's with me, not a salesperson.
Thirty minutes with a consultant
Not a salesperson. We tell you which lever fits your case, and if none does, we say that too.
Are you absent, mentioned or recommended?